Re: [Zope-dev] Re: zpatterns-0.4.0a5: problem withunregistering transactionals
"Phillip J. Eby" a écrit :
Question... Are you on 2.1.x or 2.2?
I'm still on 2.1.6. I cannot afford a migration of my product now since it is still in developpement and my customer wants it to work as soon as possible.
Also, in your example, tpc_entered is not set because tpc_entered is only true if the transaction commit process (tpc = two-phase commit) has been started. I thought that tpc_entered would go back to 0 when the transaction is over. It is clear that I don't know how the transaction machinery really works.
regards, jephte.clain@univ-reunion.fr
At 07:12 PM 7/27/00 +0400, Jephte CLAIN wrote:
"Phillip J. Eby" a écrit :
Question... Are you on 2.1.x or 2.2?
I'm still on 2.1.6. I cannot afford a migration of my product now since it is still in developpement and my customer wants it to work as soon as possible.
Okay, that makes sense. 2.1.6 has a weirdness in the ZPublisher exception hook that begins a new transaction instead of aborting the old one cleanly, and that ultimately causes the objects to get aborted twice. I reported this a month or two ago and it got fixed in 2.2. You're going to have to go with your patch, though, because there's not a simple patch you can apply to 2.1.6 to fix the problem in Zope itself. Unfortunately, it is hard for me to tell whether your patch has any other repercussions. :( I will try to review the structure of my Transactional class and see if there are any side effects (besides the one you've found) of an object being aborted twice.
Also, in your example, tpc_entered is not set because tpc_entered is only true if the transaction commit process (tpc = two-phase commit) has been started. I thought that tpc_entered would go back to 0 when the transaction is over. It is clear that I don't know how the transaction machinery really works.
It's probably because of the double-abort weirdness. :(
participants (2)
-
Jephte CLAIN -
Phillip J. Eby